Cognitive Neuroscience and Neuropsychology
NeuroReport 8, 1761–1765 (1997)
1
1
PREVIOUS work suggests that speech sounds incorporating short-duration spectral changes (such as the
formation transitions of stop consonants) rely on left
hemisphere mechanisms for their adequate processing to
a greater degree than do speech sounds incorporating
spectral changes of a longer duration (such as vowel
sounds). Ten normal subjects were scanned using
positron emission tomography while discriminating pure
tone stimuli incorporating frequency glides of either
short or long duration. A comparison of these two conditions yielded significant activation foci in left
orbitofrontal cortex, left fusiform gyrus, and right cerebellum. Because non-linguistic stimuli were used, these
foci must reflect some basic low level aspect of neural
processing that may be relevant to speech but cannot be
a consequence of accessing the speech system itself.
Key words: Auditory perception; Cerebellum; Frontal
cortex; Fusiform gyrus; Hemispheric specialization;
Language; Positron emission tomography; Speech acoustics
Introduction
1
1
1
p
Auditory language comprehension is a multistage
process relying on many different cognitive functions
as an acoustic signal is perceived, parsed into
subphonemic features, and the features synthesized
into a semantically coherent utterance. In approximately 96% of right-handers and 70% of lefthanders, the left hemisphere is preferentially involved
in the later (phonological and semantic) stages of this
process,1 and it is possible that speech processing
proceeds asymmetrically from its earliest stages.
Indeed, it has been hypothesized that the left hemispheric specialization for speech may be based, in
part, on prepotent mechanisms within this hemisphere for the processing of auditory patterns of a
fine temporal grain, such as are found in speech.2–4
Some phonetic features (such as voicing and place of
articulation) are instantiated in spectral changes of an
acoustic signal that happen with a time course of tens
of milliseconds. For example, the perception of place
of articulation contrasts (like /ba/ vs /da/) requires
accurate tracking of formation trajectories at
constant-vowel transitions which are typically no
more than 40 ms long.
Evidence for a left hemispheric specialization for
the processing of acoustic transients comes from
several sources. First, patients with left hemisphere
lesions and aphasia appear to be impaired at discriminating phonemes on the basis of voicing and place
of articulation, but such patients do not appear to
© Rapid Science Publishers
Left-hemisphere
specialization for the
processing of acoustic
transients
Ingrid S. Johnsrude,CA
Robert J. Zatorre, Brenda A. Milner
and Alan C. Evans1
Neuropsychology/Cognitive Neuroscience Unit
and 1McConnell Brain Imaging Centre,
Montreal Neurological Institute, McGill
University, 3801 University Street, Montreal,
Canada H3A 2B4
CA
Corresponding Author
show impairment of vowel discrimination, which is
made on the basis of spectral information present in
the auditory signal over a longer period.5–9 Second,
dichotic listening studies in normal populations
demonstrate a greater right ear (left hemisphere)
advantage for speech sounds that require perception
of short duration spectral changes for their accurate
identification than for speech sounds that do
not.2,3,10,11 For example, Shankweiler and StuddertKennedy2 found no differences in accuracy of
identification of steady-state vowels between stimuli
presented to the left ear and those presented to
the right. In contrast, when perception of stop
consonant–vowel syllables was examined (/pa, ta, ka,
ba, da, and ga/), they found that subjects were more
accurate at reporting stimuli presented to the right
ear, to a highly significant degree. Finally, two recent
positron emission tomography (PET) studies12,13
revealed increased activation in left hemisphere areas
in response to rapidly changing auditory cues. In a
study in which normal subjects listened passively to
speech-like syllables incorporating formation transitions, Belin and his colleagues12 found bilaterally
symmetrical activity in auditory areas when the
formation transitions were long (200 ms), but a leftward asymmetry in activation in response to short
(40 ms) formant transitions. Fiez et al.12 observed that
increases in cerebral blood flow (CBF) in the left
frontal operculum were larger when their subjects
performed an auditory detection task upon stimuli
that incorporated stop consonants than for steadyVol 8 No 7 6 May 1997
1761
I. S. Johnsrude et al.
1
11
11
11
state vowels. Thus, speech signals that are stable or
that change only slowly in time (over hundreds of
milliseconds) appear to be processed bilaterally. In
contrast, speech signals that change rapidly in time
(over tens of milliseconds) appear to be preferentially
processed by left hemisphere mechanisms.
The present PET activation study was designed to
examine the idea that providing fine-grained temporal
resolution to the auditory percept may be a general
specialization of the left hemisphere and may apply
to the processing of any sound that changes rapidly
in time, not just sounds with linguistic relevance. If
this is the case, then the processing of non-speech
pure tones that incorporate frequency glides with the
temporal characteristics of stop consonant formant
transitions ought to involve left hemisphere regions
preferentially, when compared with processing of
pure tone stimuli incorporating frequency glides of
a longer duration. Speech signals are spectrally
complex however, and the left hemispheric specialization observed in the studies described above may
not be related to the temporal structure of acoustic
transients per se, but may instead depend primarily
on their spectral characteristics. The stimuli used in
this experiment are qualitatively different from
speech stimuli in their spectral properties. It was
hoped that this would be a stronger test of the
hypothesis that the left hemisphere possesses specialized mechanisms for the processing of fine-grained
temporal structure.
Materials and Methods
The experimental protocol was approved by the
ethical review committee of the Montreal Neurological Institute and Hospital. Twelve right-handed
neurologically normal participants (six men, six
women; mean age 23 years) were recruited from the
McGill community, and trained on 80 trials of each
of the experimental tasks (see below). Any subject
who did not score at least 75% correct on both tasks
during the last 40 trials was not scanned. Two
subjects (one male, one female) were thus excluded.
Two sets of sound stimuli were synthesized on an
IBM-compatible computer. The ‘short glide’ set
consisted of eight items; 220 ms steady-state pure
tone at one of four frequencies (1250, 1600, 1950, or
2300 Hz), preceded by a 30 ms 350 Hz ascending
or descending linear frequency glide. The chosen
steady-state frequencies, the absolute frequency
difference subtended by the glide, and the duration
of the glide were all within the range observed in the
second formant of English stop consonants. The set
of ‘long glide’ stimuli was almost identical, except
that the duration of the glide component was extended to 100 ms, and the duration of the steadystate component was decreased to 150 ms, so that the
absolute duration of the stimuli was held constant at
250 ms. Similarly, the absolute frequency difference
subtended by the glide was held constant at 350 Hz.
Subjects underwent PET scans in two separate
conditions, presented in a random order. In both
conditions, subjects listened to the glide stimuli, presented in pairs, through insert earphones (EarTone
ER3A) at 75 dB SPL (A). The ‘short glide’ stimuli
were used in one condition, whilst the ‘long glide’
stimuli were used in the other. In both conditions,
subjects were asked to make same–different judgements on the pairs, indicating their choice by a key
press response with the right hand. Within each pair,
the steady-state positions of the stimuli were identical in frequency. In half the trials, however, the glide
11
11
1p
FIG. 1.
Schematic drawing of the scan time course and sample stimuli used in the experiment. Two trials are shown in each condition.
1762 Vol 8 No 7 6 May 1997
The processing of acoustic transients
1
1
1
FIG. 2. Averaged PET t-statistic volume superimposed upon the averaged MR image. The range of t-values for the PET data is coded by the
colour scale. (A) Sagittal section, taken 26 mm to the left of midline. (B) Horizontal section taken 30 mm below the bicommissural plane.
(C) Horizontal section, taken 18 mm below the bicommissural plane. (D) Horizontal section taken at 10 mm above the bicommissural plane.
(E) Horizontal section, taken 13 mm below the bicommissural plane. Subtraction of the regional cerebral blood flow observed in the long
glide condition from that observed in the short glide condition yielded significant focal changes in the left orbitofrontal region; area 11
(stereotaxic coordinates x = –31, y = 44, z = –19; t = 4.11; shown on A and C); in the left fusiform gyrus; area 19/37 (x = –32, y = –50, z = –12;
t = 3.55; as shown in A and E); and the right cerebellum (x = –28, y = –61, z = –29; t = 3.54; shown on B). In addition, an activation focus was
observed in the left middle occipital gyrus, area 18 (x = –24, y = –85, z = 9, shown on A and D), although this focus did not reach significance (t = 3.25).
1
1
p
component was different (i.e. ascending to the steadystate tone in one item, and descending to that tone
in the other item). Pairs were presented with a 3 s
inter-trial interval, and the interval between items
of a pair was 1 s (Fig. 1). Subjects were scanned for
60 s, and each task commenced approximately 40 s
before scanning, and continued for 45 s after the scan
was completed (32 pairs presented).
PET data were acquired using a Scanditronix
PC-2048 system, which produces 15 slices at an
intrinsic resolution of 5.0 ´ 5.0 ´ 6.0 mm. Using the
bolus H216O method without blood sampling, relative distribution of CBF was measured in the two
conditions. PET images were reconstructed using an
18 mm full-width half-maximum filter and normalized for global CBF value. A high resolution MR
image (160 slices, 1 mm thick) was also obtained for
each subject (Philips Gyroscan 1.5T) and each pair
of MR and PET data sets was resampled into a standardized stereotaxic coordinate system.14,15 The MR
and PET images were averaged across subjects for
each activation state to obtain an average MRI and a
mean CBF change image. Significant focal CBF
changes were identified in the averaged PET volume
by converting it to a t-statistic volume (by dividing
each voxel by the mean standard deviation in normalized CBF for all intracerebral voxels).16 The functional images were merged with the averaged MR
image to allow localization of significant t-statistic
peaks, which were identified by application of an
intensity threshold to the t-statistic images.
The presence of significant focal changes was
tested by a method based on three-dimensional
Gaussian random-field theory.16 Values of t > 3.5
were deemed statistically significant (p < 0.0002, onetailed, uncorrected).
Vol 8 No 7 6 May 1997
1763
I. S. Johnsrude et al.
1
11
Results
Subjects were significantly more accurate at making
same–different judgements on long glide pairs than
on short glide pairs (t = 4.72; p < 0.001), although
both tasks were performed well (mean 95% correct
for the long glide task and 86% correct for the
short glide task). There was no difference in mean
response latency in the two tasks (t = 1.84; p > 0.05;
mean long glide latency 893 ms, mean short glide
latency 973 ms).
When the pattern of CBF observed in the long
glide condition was subtracted from that in the short
glide condition, significant cortical activations were
observed in only three regions: the left orbitofrontal
cortex, fusiform gyrus, and in the right cerebellar
hemisphere, pars lateralis (Fig. 2). A CBF decrease
was observed in right mid-dorsolateral frontal cortex
(area 9; x = 27, y = 3, z = 36; t = 3.50).
11
Discussion
11
11
11
1p
All of the neocortical activation foci were in the left
hemisphere. The right lateral cerebellar activation is
consistent with these left neocortical foci: the right
lateral cerebellum is richly interconnected with the
left cerebral hemisphere via the red and pontine nuclei
and the thalamus, and several investigators have
posited a role for the right cerebellar hemisphere in
at least some aspects of language function.17,18 Given
that the task demands of the two conditions were
very similar, differing only in the duration of the
glide component of the sounds, our findings support
the contention that are specialized mechanisms for
the processing of acoustic transients in the left cerebral hemisphere. Since we used pure tone stimuli of
no linguistic value, we conclude that this specialization, while relevant to speech perception, must reflect
a more general functional property of the human left
hemisphere, and may apply to the processing of any
short duration spectral change.
The fusiform gyrus activation was particularly
interesting, given that studies of phonemic monitoring consistently show activation in this region.19,20
For example, Zatorre and his colleagues20 performed
a PET study in which subjects listened to pairs of
monosyllabic English words. When the subjects were
asked to monitor the pairs for the occurrence of a
particular stop-consonant, and activation in this
condition was compared with that obtained when
they were simply listening to the words, significant
CBF change was observed in area 19/37 (stereotaxic
coordinates x = –34, y = –57, z = –9), about 8 mm
posterior to the focus observed in the current study.
Similarly, Démonet and his colleagues19 asked subjects to perform a stop consonant detection task that
1764 Vol 8 No 7 6 May 1997
had been made perceptually difficult. When activation in this condition was compared with that
obtained during performance of a phoneme detection
task that was perceptually simpler, the only significant CBF change observed was in the left area 19
(stereotaxic coordinates x = –32, y = –74, z = –12),
about 2.4 cm posterior to that observed in the present
study. Thus, the posterior fusiform region in the left
hemisphere may be involved in processing the
acoustic manifestations of some phonetic features, as
would be necessary for the correct identification of
a target stop consonant.
Given that electrophysiological and neuromagnetic
studies in cats, monkeys and humans have demonstrated the existence of neurones in the region of
primary auditory cortex that are apparently able to
preserve the temporal structure of transient auditory
signals, and to encode the direction of a transient
change,21–25 we expected to see some differences in
activation in the region of auditory cortex. This was
not the case. Either the two types of stimuli used in
this study are processed similarly at early stages, or,
as seems more likely, the sensitivity of the PET
subtractive technique in combination with the
discriminative task we used is insufficient to distinguish between the relevant auditory systems. We are
currently exploring alternative PET methods which
may be more sensitive to blood flow changes in
this region.
Conclusion
The present study examined the brain regions
involved in processing a single property of speech
sounds that typically yield a strong right ear advantages in dichotic listening studies; namely, the
temporal structure of stop consonant formant transitions. We did not attempt to respect the spectral
properties of such formant transitions, and it is
important to note that the speech signal is full of
other information at changes more slowly over time.
Data from this experiment, however, do provide
support for the hypothesis that the left hemisphere
possesses specialized mechanisms for the processing
of acoustic information of a fine temporal grain.
References
1. Branch C, Milner B and Rasmussen T. J Neurosurg 21, 399–405 (1964).
2. Shankweiler D and Studdert-Kennedy M. Q J Exp Psychol 19, 59–63 (1967).
3. Studdert-Kennedy M and Shankweiler D. J Acoust Soc Am 48, 579–594
(1970).
4. Schwartz J and Tallal P. Science 207, 1380–1381 (1980).
5. Blumstein SE, Cooper WE, Zurif EB and Caramazza A. Neuropsychologia
15, 371–383 (1977).
6. Gow DW Jr and Caplan D. Brain Lang 52, 386–407 (1996).
7. Miceli C. Caltagirone C, Gainotti G and Payer–Rigo P. Brain Lang 6, 47-51
(1978).
8. Oscar-Berman M, Zurif EB and Blumstein SE. Brain Lang 2, 345–355 (1975).
9. Tyler L. Spoken Language Comprehension: An Experimental Approach to
Disordered and Normal Processing. Cambridge: MIT Press, 1992.
The processing of acoustic transients
1
10. Berlin CI and McNeil MR. Dichotic listening. In: Lass N, ed. Contemporary
Issues in Experimental Phonetics. New York, Academic Press, 1978: 327–374.
11. Cutting JE. Percept Psychophy 16, 601–612 (1974).
12. Belin P, Zilbovicius M, Crozier S et al. Soc Neur Abst 22, 1698 (1996).
13. Fiez JA, Raichle ME, Miezin FM et al. J Cogn Neurosci 7, 357–375 (1995).
14. Talairach J and Tournoux P. Co-planar Stereotaxic Atlas of the Human
Brain. Three-dimensional Proportional System: An Approach to Cerebral
Imaging. Stuttgart: Thieme, 1988.
15. Colins DL, Neelin P, Peters TM and Evans AC. J Comput Assist Tomogr
18, 192–205 (1994).
16. Worsley KJ, Evans AC, Marrett S and Neelin P. J Cerebr Blood Flow Metab
12, 900–918 (1992).
17. Fiez JA. Neuron 16, 13–15 (1996).
18. Leiner HC, Leiner AL and Dow RS. Trends Neurosci 16, 444–447 (1993).
19. Démonet J-F, Price C, Wise RJS and Frackowiak RSJ. Brain 117, 671–682
(1994).
20. Zatorre RJ, Meyer E, Gjedde A and Evans AC. Cerebr Cortex 6, 21–30 (1996).
Glass I and Wollberg Z. Hearing Res 9, 27–33 (1983).
Kaukoranta E, Hari R and Lounasmaa O. Exp Brain Res 69, 19–23 (1987).
Phillips D. J Exp Psychol: Human Percept Perform 19, 203–216 (1993).
Steinschneider M, Schroeder C, Arezzo J and Vaughan H Jr. Electroencephalogr Clin Neurophysiol 92, 30–43 (1994).
25. Whitfield I and Evans E. J Neurophysiol 28, 655–681 (1965).
21.
22.
23.
24.
ACKNOWLEDGEMENTS: We thank the staff of the McConnell Brain Imaging
Centre and Pierre Ahad for their technical assistance. Supported by grants from
the Medical Research Council of Canada (SP-30, MT-2624 and MT-11541), and
from the McDonnell-Pew Program in Cognitive Neuroscience.
Received 21 February 1997;
accepted 6 March 1997
General Summary
The identification of a variety of speech sounds depends on the perception of changes in the sound spectrum that happen over tens
of milliseconds, such as occurs in the transition of the second formant in stop-consonants. In order to determine whether the lefthemispheric specialization for language may be based in part on a relative specialization of this hemisphere for the processing of
acoustic transients, we performed a PET study in which 10 normal subjects were scanned while discriminating pure tone stimuli
incorporating frequency glides of either short, or long duration. A comparison of these two conditions yielded significant blood flow
changes in left cortical areas and in right cerebellum. Because non-linguistic stimuli were used, we concluded that this asymmetrical activation (while relevant to speech perception) must reflect a more general specialization of the human left hemisphere that
applies to the processing of any short duration spectral change.
1
1
1
1
p
Vol 8 No 7 6 May 1997
1765